Data Science Techniques for Unbiased & Efficient Production Analysis

Abstract

Analyses have been widely applied in production forecasting of oil and gas production in both conventional and unconventional reservoirs. In order to forecast production, to estimate reservoir properties, or to evaluate resources, various statistical and machine learning approaches have been applied to various reservoir analysis methods. Nevertheless, many of these methods are suboptimal in detecting production trends in different wells due to data artifacts (noise, data scatter and outliers, inadequate SCADA systems, production allocation problems) that obscure unit reservoir signals, production trends, and more leading to large forecast error, or fail due to lack of data access (inadequate SCADA systems, missing or abhorrent data, and production allocation problems). This work outlines a method that is currently being used in a commercial setting which combines advanced analytics and machine learning with a modern cloud architecture, provide rapid, repeatable, unbiased estimates of original hydrocarbon -in-place (OHIP), estimated ultimate recovery (EUR), and remaining recoverable (RR), and even deliverability forecasts - all in the presence of abhorrent data.